KiviDB posts 21.2M ops/sec Redis-compatible benchmark

KiviDB’s creator published a Hacker News post on Oct. 19 saying the Rust database reached 21.2 million operations a second in a pipelined GET benchmark against Redis-compatible systems.

The self-reported result matters because KiviDB targets Redis and Valkey workloads, claiming about 98% to 99% support for Redis 8.10 commands, while Redis reached 2.4 million operations a second in the same pipelined test.

The project’s GitHub repository describes KiviDB as an AGPLv3-licensed in-memory database written in Rust, with support for strings, hashes, lists, sets, sorted sets, streams, JSON, time series, vector sets, full-text and vector search, Lua scripting, access controls, persistence, replication, Redis Cluster-compatible sharding and TLS.

The creator said the benchmark used two AWS c7gn.metal servers in the same placement group, 10 million keys of 100 bytes each, 32 cores and the memtier-benchmark tool. In the pipelined GET test, KiviDB’s 21.2 million operations a second compared with 9.1 million for Dragonfly 2.0, 4.7 million for a 32-shard Redis Cluster and 2.4 million for Redis 8.10, according to the post.

The comparison is not one-way. Without pipelining, the post said Dragonfly reached 3.35 million operations a second, ahead of KiviDB at 3.07 million. It also said KiviDB used 177 bytes of memory per key, compared with 134 bytes for Dragonfly and 178 bytes for Redis.

KiviDB supports RESP2 and RESP3, versions of the Redis Serialization Protocol used by Redis clients. On Linux, the project uses io_uring, a kernel interface for asynchronous input and output, with a Tokio fallback for other operating systems.

The benchmark and compatibility claims are published by the project’s creator and repository, not an independent test lab. The creator said some Redis-command issues remain and pointed users to the project’s GitHub issue tracker.

Source: Hacker News post and GitHub repository, KiviDB, Oct. 19, 2026